Band Structure of the Four Pentacene Polymorphs and Effect on the

Thorsten Vehoff , Björn Baumeier , Alessandro Troisi and Denis Andrienko ..... Simulation of coherent two-dimensional electronic spectroscopy by Davy...
1 downloads 0 Views 497KB Size
J. Phys. Chem. B 2005, 109, 1849-1856

1849

Band Structure of the Four Pentacene Polymorphs and Effect on the Hole Mobility at Low Temperature Alessandro Troisi* and Giorgio Orlandi Dipartimento di Chimica “G. Ciamician”, UniVersita´ di Bologna, Via F. Selmi 2, 40126 Bologna, Italy ReceiVed: September 18, 2004; In Final Form: NoVember 24, 2004

The band structure of the four known polymorphs of pentacene is computed from first principles using the accurate molecular orbitals of the isolated molecule as the basis for the calculation of the crystalline orbitals. The computed bands are remarkably different for each polymorph, but their diversity can be easily rationalized using a simple analytical model that employs only three parameters. The effect of the electronic structure on the hole mobility was evaluated using a simple model based on the constant relaxation time approximation. It is found that the mobility tensor is highly anisotropic for three of the four considered polymorphs. The practical implication of this prediction on the technology of thin-film organic transistors is discussed.

1. Introduction The possibility of practical applications for organic semiconductors, demonstrated in the late 1980s1 and followed by an impressive improvement of the performance and efficiency of the devices based on such materials,2 have renewed the interest of many researchers toward this field whose first contributions appeared more than fifty years ago.3 Organic materials (crystals or polymers) based on polyacenes, polythiophenes, and polyetilene have been used to realize lightemitting diodes (LEDs), thin-film transistors (TFTs), and photovoltaic cells, and an increasingly large set of data on these systems are now available.4-8 The most important property of these materials is the charge carrier mobility µ, whose lower limit for practical application is 100 cm2 V-1 s-1. Organic synthesis provides, in principle, the possibility to fine-tune the charge-transport properties, but the mobility of these materials is very difficult to predict, and the available mobility data lack a proper rationalization. Many groups are therefore active in the development of phenomenological theories9 and computational models10-13to provide reliable predictive and interpretative tools. In ordered organic materials such as pentacene,2 the lowtemperature transport is described as band-like (i.e., delocalized carriers move coherently across the crystal and are scattered by the lattice phonons). This mechanism is characterized by a power law dependence of the mobility upon the temperature (µ ≈ T-n).14-16 At higher temperatures (∼300 K), polaron transport becomes important; that is, the charge carriers (and their associated lattice deformation) move by thermally activated hopping leading to an Arrhenius-type temperature dependence of the mobility (µ ≈ exp(-Ea/kT)).17,18 The possibility of a unified description of both transport regimes through suitable effective Hamiltonians is discussed by several authors.9a,19-21 While phenomenological theories account qualitatively for the observations, they cannot explain the differences observed for similar materials, for which accurate electronic structure calculations are needed. Analogous materials can show dramatic differences in the transport properties,22 because the interaction between molecules strongly and subtly depends on structural details. We illustrate this point in the present paper, computing

the band structure of the four polymorphs of pentacene and discussing the effect of pentacene polymorphism on the lowtemperature hole mobility. Since the first reports of high hole mobility for the pentacene single crystal,23 several groups studied this24-28 and related materials29-31 for their potential application in organic electronics. Several pentacene polymorphs were grown as thin films,33-38 and one of the thin-film structures was shown to coincide35 with the bulk single-crystal structure reported by two recent studies.32,35 A classification and a rationalization was proposed by Mattheus et al.33 that also found the conditions to reproducibly grow thin films of four crystal forms.34 The possibility of different transport properties for different growth conditions is of great technological interest. In fact, one of the typical experimental setups involves pentacene thin films, grown on a silicon oxide surface between the source and drain electrodes, forming a prototype of organic thin-film transistors (OTFTs). To compute the band structure, we propose a first-principles method that uses accurate molecular orbitals computed for the isolated molecule as basis functions for the crystal wave function. A simple analytical model will be proposed to interpret all the band structure, and the results will be further justified through the use of simple orbital overlap arguments. We will not consider in this paper the polaronic mechanism of conduction, limiting our discussion to band-like low-temperature conduction. According to recent measurements,16 the band-like transport is the dominant one in pentacene up to ca. 300 K. 2. Method The most commonly used packages39 for the first-principles computation of band structures have been optimized for the calculation of materials with a relatively small number of atoms in the unit cell and with band gaps ranging from several electronvolts to zero (metals). Molecular crystals contain up to hundreds of atoms per unit cell, and they are usually insulators or semiconductors. The appropriate description of molecular orbitals (MOs) for the isolated molecule requires a split-valence atomic basis set with the inclusion of polarization functions. This basis set makes the calculation of the molecular crystal band structure extremely heavy and the convergence of a self-

10.1021/jp0457489 CCC: $30.25 © 2005 American Chemical Society Published on Web 01/19/2005

1850 J. Phys. Chem. B, Vol. 109, No. 5, 2005

Troisi and Orlandi

consistent calculation quite problematic. However, in the vast majority of the cases, the electronic coupling between MOs localized on different molecules is much smaller (∼1.5 eV). Under this condition, the localized MOs provide a natural basis for the calculation of the molecular crystal band structure. In the following, we describe the computational method after reviewing, for completeness, the general equations for the calculation of the band structure on a localized basis. Background. Let {φR} be a set of localized one-electron wave functions in an elementary cell (e.g., atomic orbitals or a linear combination of them). The Bloch orbitals (i.e., the basis functions adapted for the translational symmetry) are39

ψkR )

1

xN

∑T φR(r - T) exp(ikT)

(1)

where T is an element of the direct lattice with unit vectors b a, B b, and b c.

a + nbB b + ncb c T ) n ab

(na, nb, nc integers)

(2)

and k is a vector of the reciprocal space (unit vectors b a*, B b*, and b c*):

a * + xbB b * + xcb c* k ) x ab

(xa, xb, xc real)

(3)

The matrix elements of the effective one-electron Hamiltonian are

〈ψkR|H|ψkβ〉 )

)

1 N

1 N

〈∑

||

φR(r - T′) exp(ikT ′) H

T′

∑T φβ(r - T) exp(ikT)



(4)

∑ ∑ exp[ik(T-T′)]〈φR(r-T′)|H|φβ(r-T)〉 T′ T

(5)

The elements in the sum depend on the T - T′ difference, and because

∑ ∑ f(T - T′) ) N∑T f(T) T′ T

(6)

we get

〈ψkR|H|ψkβ〉 )

∑T exp(ikT)〈φR(r)|H|φβ(r - T)〉

(7)

The relevant matrix elements are conveniently labeled as

VRβT ) 〈φR(r)|H|φβ(r - T)〉

(8)

In a molecular crystal, the {φR} set can be defined by the molecular orbitals of the isolated molecule(s) (one set of orbitals for each molecule in the elementary cell). Because the {φR} are localized, one can make a (short) list of nonzero VRβT elements, which can be determined on a purely geometrical basis. The Hamiltonian matrix elements are a summation, with a phase factor, over the list of nonzero couplings: k ) 〈ψkR|H|ψkβ〉 ) HRβ

∑T VRβT exp[i2π(xai + xbj + xck)]

(9)

where we expressed T and k in terms of their components (eqs 2-3). Because the {φR} are non-orthogonal, the overlap matrix elements should be computed as in eqs 8-9: k SRβ )

∑T SRβT exp[i2π(xai + xbj + xck)] SRβT ) 〈φR(r)|φβ(r - T)〉

(10) (11)

The band energies for each k value are the solutions of the generalized eigenvalues equation:

HkCk ) SkCkk

(12)

Implementation. For a given crystal structure with m molecules in the elementary cell, the MOs for the different isolated molecules are computed. These orbital energies are the elements VRR(T)0).40 We used the B3LYP hybrid functional for this computation, but any other method (Hartree-Fock, HF, or density functional theory, DFT) based on the one-electron effective Hamiltonian could be applied. An orbital window around the Fermi energy is selected, because usually few higherenergy occupied orbitals and lower-energy virtual orbitals are sufficient to describe the transport properties (the interaction between selected and excluded orbitals is neglected). A list of couples of interacting molecules (molecules whose orbitals interact) has been compiled. In our case, we include the interaction between two molecules if the closest distance between their atoms is smaller than a cutoff distance of 5.5 Å. Each couple of molecules is identified by the five-number code: m1, m2, [na, nb, nc]. m1 and m2 identify the molecules in the elementary cell; na, nb, nc identify the translation of the molecule m2 with respect to m1. For example, 1 2 [0 0 0] is the coupling between molecules 1 and 2 in the elementary cell, and 1 1 [1 0 0] is the coupling between molecule 1 and its image translated by 1 unit vector a. The couplings matrix elements VRβT and the overlaps SRβT are computed between all the considered orbitals on the interacting molecules according to eqs 8 and 11. The Hamiltonian is the same used for the isolated molecule calculation. Because one-electron Hamiltonians depend on the density matrix, one of the greatest difficulties in the calculation of the band structure of metals or small-gap semiconductors is the calculation of the total electron density through integration over the reciprocal space.39 For a molecular crystal, this step is not necessary, because, to an excellent degree of approximation, the density matrix of the isolated molecule can be used to build the one-electron Hamiltonian for the crystal. Therefore, the electron density is computed self-consistently for the isolated molecule and used to build a tight binding-like Hamiltonian for the crystal. A grid of points {ki} for which the crystal orbitals have to be computed is selected. For each value ki, the matrices Hk and Sk are built, and eq 12 is solved for Ck and k. The size of the algebraic problem is given by the total number of occupied and virtual orbitals included in the calculation. Other Tight Binding Schemes. Differently than in other tight binding schemes (EH,41 DFTB42) that compute the tight binding interactions between all the atom pairs, in this approach the strong intramolecular interactions are treated at a high level of accuracy, while the weaker intermolecular interactions are approximated perturbatively. It is therefore possible to arbitrarily improve the description of the MOs by increasing the atomic basis set without increasing the size of the crystal Hamiltonian (eq 12) that always contains the same subset of MOs.

Band Structure of Pentacene Polymorphs A similar approach was used recently by Cheng et al.11 that employed a semiempirical Hamiltonian to compute the coupling. An important difference is that we include in this treatment all the couplings between the selected subset of orbitals and not only the coupling between degenerate orbitals. These additional matrix elements could be particularly important in the presence of quasi-degenerate molecular orbitals or in molecular crystals with different molecules in the elementary cell. Crystal Geometry. We used the unit cell parameters of the four polymorphs as they are listed in Table 2 of ref 33. We also refer to this paper for a description of the crystal structure and a possible explanation for the polymorphism. Here, we identify the polymorphs as I, II, III, and IV corresponding to a distance d(001) between their ab planes of 14.1, 14.4, 15.0, and 15.4 Å. The experimental determination of their cell parameters is described in ref 35 for polymorph I and ref 34 for polymorphs II, III, and IV. Polymorph I is the commonly adopted structure in single crystals, and its detailed structure is available from X-ray diffractometry.35,32 As in other polyacenes, two nonequivalent molecules per unit cell are arranged in herringbone fashion with space group symmetry P1h. The other polymorphs adopt a similar structure, but the detailed crystal geometry was not determined. To make our data set uniform, we performed geometry optimizations of all four polymorphs with the MM343 force field, and these geometries were used for the calculations presented in the next section. Many crystal structures of aromatic hydrocarbons were considered in the parametrization of this force field that is therefore expected to also perform well in our case. To further test the approximation of the method, we also performed geometry optimization and band structure calculation using the approximate DFT approach implemented in the code SIESTA.44 The results, not presented in this paper, indicate that this completely different approach leads to quantitatively similar results for the geometry and the band structure. It is likely that the rigidity of the molecule and the constraints of the unit cell parameters lead to a relatively sharp potential energy minimum that is easily predicted by the various methods.45 3. Results and Discussion Figure 1 shows the results for a band structure calculation of polymorph I of pentacene. All ab initio bands are computed using the B3LYP hybrid density functional and the 6-31G(d) basis set. A recent systematic study by Huang and Kertesz46 shows that this basis set is adequate for the evaluation of the intermolecular coupling and that the error due to the incompleteness of the basis set is